Topic
3 replies Latest Post - ‏2013-04-18T05:54:23Z by dlmcnabb
tluther
tluther
4 Posts
ACCEPTED ANSWER

Pinned topic User Quota management for HSM

‏2013-04-16T07:20:52Z |

Hi

I got a question in regard to GPFS quota management with a SONAS system and HSM. I found in the forum that HSM migrated files are not counted anymore for the GPFS User Quota (stub file size = 0 bytes).

Considering regular HSM migration of older/larger files to tape would mean that the user's free space increases with HSM migration. So basically HSM migration would free up user /group space that is limited by quotas.

Now the answer which I did not find yet is what happens if files are recalled into the disk pool. I assume this is charged again to User/group quotas and suddenly those users/groups cannot create further new data since the recall filled up their quota?

In that case, what happens during the recall of a large file which exceeds the quota, will that file be completely recalled (in SONAS recall happens as root user)?

What happens during file access that would trigger recalls if the quota is already exceeded? Read error or recall (as root) and even further exceeding the quota for the owning user/group but read will work?

 

  • dlmcnabb
    dlmcnabb
    1012 Posts
    ACCEPTED ANSWER

    Re: User Quota management for HSM

    ‏2013-04-16T16:26:25Z  in response to tluther

    Since the recall function is done on the HSM session node with a root id (it has no idea who the initiator of the recall is), the recall will succeed because quota is not enforced against a root process. The user/group quota is adjusted by the space used, just with no enforcement. If this goes over any quota, the user/group will not be able to allocate any more space on their own until some files are deleted or migrated to HSM.

    • tluther
      tluther
      4 Posts
      ACCEPTED ANSWER

      Re: User Quota management for HSM

      ‏2013-04-17T09:22:02Z  in response to dlmcnabb

      Hi Dave

      Thanks for the answer. This is what I though and I could also validate that in meantime.

      It brings however other questions up:

      1) If it is a fileset quota, does GPFS also just adjusted the space used, without enforcement? Since there it probably doesn't matter whether it is a root process?

      2) How does GPFS enforce quotas if existing files are just increased by users? Will GPFS allow file increase beyond soft/hard quotas or fail on write of such file changes/increases?

      If quotas are just enforced for new files, it could explain that recalls will always work, no matter which user will do it or which kind of quota might be hit first. Since recall is basically an increase of an existing 0byte file in the disk space.

      • dlmcnabb
        dlmcnabb
        1012 Posts
        ACCEPTED ANSWER

        Re: User Quota management for HSM

        ‏2013-04-18T05:54:23Z  in response to tluther

        1) Fileset quota will be accounted just like user/group quotas. However, there is a configuration variable that will enforce fileset quota limits on root: If enforcement is on, then the HSM restore will fail when the fileset quota is exceeded. Only part of the file will be restored if there was some room available when it started.

           mmchconfig enforceFilesetQuotaOnRoot=yes -i

        This can be turned on/off dynamically.

        2) When a user tries to allocate space on a write, the application will get EDQUOT if it would go over the hard limit. You can go past the soft limit for a short time depending on the grace period setting. If the grace period has past without dropping below the soft limit, then the next write that allocates space will get EDQUOT.